MIT releases AI model designed to detect Alzheimer's disease years before symptoms appear
Researchers at MIT have released FINGERS-7B, an AI model trained to identify people at risk for Alzheimer's disease in the preclinical stage, when the disease is active but memory loss has not yet begun. The model combines lifestyle, clinical, genomic, and proteomic data from tens of thousands of at-risk individuals to spot disease markers that no single data source can detect alone.
The gap between when Alzheimer's disease begins and when symptoms appear can stretch a decade or longer. That window is where intervention becomes possible.
How the model works
FINGERS-7B was trained to learn jointly from multiple data types: lifestyle information, clinical records, biomarkers, genomic data, and proteomic signals. The broader platform, called FINGERPRINT, pairs the model with AI agents that run automated analyses across these domains.
The core principle is that disease risk becomes easier to detect when many layers of biology and behavior are examined together. The model searches for patterns across all data types simultaneously, rather than treating each source separately.
"Each of us carries a biological fingerprint-a unique combination of signals that reveal disease risk," said Adrian Noriega, an MIT-Novo Nordisk AI Fellow and FINGERPRINT co-lead. "FINGERPRINT is a discovery acceleration engine composed of specialized agents and new foundation models that interpret these biological signals to help us find novel biomarkers, prevention interventions, and therapeutics."
Performance on clinical data
On datasets from the WW-FINGERS network, FINGERS-7B delivered four times more accurate preclinical diagnosis than prior methods. The model also showed a 130 percent improvement in responder stratification-a measure of how well researchers can identify which people are likely to benefit from specific interventions.
Preclinical Alzheimer's is the stage before cognitive symptoms appear. Earlier risk identification could expand the window for testing prevention strategies, lifestyle changes, or therapies before significant decline begins.
The model generates individualized analyses for each person. It estimates risk, forecasts the likely course of cognitive decline, and predicts the effect of possible interventions ranging from dietary changes to drug treatments.
Building on existing prevention research
The work builds on the FINGER study, a landmark effort focused on cognitively unimpaired but at-risk older adults. That study inspired the WW-FINGERS network, which now spans 40 countries and includes 30,000 participants.
The FINGERPRINT project layers additional biomarker, genomic, and proteomic datasets on top of the WW-FINGERS clinical and lifestyle data. MIT's Aging Brain Initiative provided a $100,000 grant last June to launch the effort. Within 10 months, the team trained the model, deployed it, and made it available for outside use.
Open access and shared infrastructure
The team has made the model open source. Its weights, training code, and evaluation pipelines are public, allowing other research groups to apply FINGERS-7B to their own patient cohorts and contribute results back.
The model is deployed in the AD Workbench, a secure cloud environment operated by the Alzheimer's Disease Data Initiative and used by Alzheimer's researchers worldwide. Researchers can use the system without moving sensitive patient data or building new infrastructure.
Industry partners include Alamar Biosciences and Novo Nordisk. Institutional partners include the Broad Institute, Yale University, Imperial College London, and Brigham and Women's Hospital.
Before the public release, the project attracted outside attention. In February, the Davos Alzheimer's Collaborative and the FINGERS Brain Health Institute announced a partnership to use FINGERPRINT to support prevention research, with a stated goal of making the work globally inclusive.
Practical implications for research
Earlier risk prediction could help researchers identify people for prevention studies before symptoms appear, when interventions may have a better chance of changing disease trajectory.
The responder stratification results point to a second use: sorting people into more precise groups for trials and treatment research. In a field where Alzheimer's risk is shaped by biology, health history, and everyday life, that sorting could make studies more targeted.
The open-source release also matters. Because outside groups can test the model on their own cohorts through shared infrastructure, the project functions as shared research infrastructure rather than a closed product. That structure may speed validation, widen participation, and push Alzheimer's prevention research toward a more integrated, data-rich approach.
Learn more about AI for Science & Research and AI Research Courses to understand how AI is being applied to medical discovery and prevention.
Your membership also unlocks:
